TOO MUCH OF A GOOD THING? NEGATIVE EFFECTS OF HIGH TRUST AND

TOO MUCH OF A GOOD THING? NEGATIVE EFFECTS OFHIGH TRUST AND INDIVIDUAL AUTONOMY IN

SELF-MANAGING TEAMS

CLAUS W. LANGFREDWashington University

A high level of trust can make the members of self-managing work teams reluctant tomonitor one another. If low monitoring combines with high individual autonomy, teamperformance can suffer. Data from 71 self-managing teams of MBA students demon-strated this effect. High trust was associated with higher team performance whenindividual autonomy was low but with lower performance when individual autonomywas high. Additional analysis showed a moderated mediating role of monitoring andautonomy in the relationship between trust and performance.

Doveryai, no proveryai. (“Trust but verify.”)

- Old Russian proverb

Trust is commonly accepted as having a varietyof positive effects (Kramer, 1999; Kramer & Tyler,1996), and research has focused on exploring thesebenefits. In every one of 43 empirical studies re-viewed by Dirks and Ferrin (2001), trust was ex-pected to result in benefits to individuals or theirorganizations. These expected benefits includedimproved communication, more organizational cit-izenship behaviors, less competitive behavior innegotiations, higher group performance, less con-flict, and greater job satisfaction. However, whilethe effects of trust on attitudes and perceptionshave been found to be fairly consistent and posi-tive, its effects on behavior and performance havebeen “weaker and less consistent” (Dirks & Ferrin,2001: 455), suggesting that trust might exert a mod-erating rather than a direct effect on performanceoutcomes.

The lack of “main effects” is particularly pro-nounced in teams and groups. Neither Friedlander(1970), Kegan and Rubinstein (1973)), nor Dirks(1999) could demonstrate any significant effect ofintragroup trust on group performance. Only mod-erating (or interactive) effects of trust, not directeffects, were significant in two studies (Dirks, 1999;Simons & Peterson, 2000), and only Dirks (1999) fo-cused directly on team performance as an outcome.

In this study, I explored how trust and monitor-

ing interacted with individual autonomy to affectperformance in self-managing teams. I depart fromthe tradition of searching for positive effects byexamining a more counterintuitive situation inwhich higher levels of trust might be actually asso-ciated with lower team performance. It was not myintent to suggest that trust does not have benefits inteams, but rather to explore boundary conditionsand contingencies under which high trust in teamscan be harmful. I suggest that self-managing teamswith high levels of individual autonomy will performbetter when trust is lower than when trust is high.

Self-managing teams, defined as groups of inter-dependent individuals that can self-regulate on rel-atively whole tasks (Cohen, Ledford, & Spreitzer,1996), present a setting in which issues of trust andautonomy are of primary importance. At the heartof a self-managing team is the discretion teammembers have in deciding how to carry out tasksand allocate work within the team. This discretionincludes decisions that have traditionally beenmade at managerial levels (Wellins, Wilson, Katz,Laughlin, Day, & Price, 1990), such as how muchautonomy to give to different team members andhow much to monitor them. Autonomy is definedas the amount of freedom and discretion an indi-vidual has in carrying out assigned tasks (Hack-man, 1983). I believe that the interaction betweenthe amount of autonomy and the amount of moni-toring in a team reveals a negative effect of trust onteam performance. Figure 1 illustrates my overallmodel.

In exploring both team- and individual-level con-cepts, it is important to specify the level of analysis.In this study, the team was the focal unit (Rous-seau, 1985), or level of theory (Klein, Dansereau, &Hall, 1994). I expected that their team could affectindividuals; for example, social pressure might in-

I am very grateful for the comments and advice of theGOMERS research group at Washington University (BillBottom, Stuart Bunderson, Kurt Dirks, Chris Long, JudiMcLean Parks, and Raymond Sparrowe), as well asNeta Moye, Gretchen Spreitzer, and three outstandingreviewers.

� Academy of Management Journal2004, Vol. 47, No. 3, 385–399.

385

fluence individual decisions on how much to mon-itor other team members. I also expected that ag-gregated individual effects could affect a team—forexample, the average levels of monitoring and ofindividual autonomy in a team might influenceteam performance.

In organizations, management decisions abouthow much to monitor individual employees are inpart (or in whole) based on trust. Trust is defined asa willingness to be vulnerable to the actions ofanother party (Mayer, Davis, & Schoorman, 1995)in situations involving some risk (Deutsch, 1958),factors such as benevolence, honesty, and compe-tence are often perceived as indicative of trustwor-thiness. In general, the more trustworthy employ-ees are perceived to be, the less an organization willmonitor them (Creed & Miles, 1996), and the lesstrust in its employees an organization has, the moremanagement feels it must monitor (Bromiley &Cummings, 1995). In a team setting, monitoring isdefined as the team members’ surveillance andawareness of other team members’ activities, andintrateam trust is the aggregate perception of trust-worthiness that team members have about one an-other. In a team, as in an organization, an inverserelationship between trust and monitoring is ex-pected to operate. High levels of intrateam trust areexpected to be associated with low levels of moni-toring, and low levels of trust are expected to beassociated with high levels of monitoring.

In self-managing teams, however, where teammembers are responsible for the monitoring, mem-bers may choose to not monitor one another whenthe level of trust is high. In such teams, socialforces may make it difficult for individual teammembers to suggest the necessity of monitoring oneanother once a high level of trust is established.

Because of the perception that monitoring resultsfrom a lack of trust, such a suggestion could beperceived as critical of other team members, andthus risks sanction and rejection by the team (Feld-man, 1984). Teams in general, and particularlythose high in characteristics like cohesiveness andtrust, can exert a powerful influence on individualsto conform (Baron, Vandello, & Brunsman, 1996),and such teams are also especially susceptible togroup decision biases like “groupthink” (Janis,1982). The generally negative connotations of mon-itoring and surveillance and the negative effectthey can have on individual motivation (Enzle &Anderson, 1993) are expected to discourage indi-vidual team members from suggesting monitoring.An individual team member might also be con-cerned that a suggestion to monitor fellow teammembers could be perceived as a violation of trustitself, leading to anger, hurt, and fear on the part ofthe team members who perceive a violation(Lewicki & Bunker, 1996). This argument suggests anumber of dynamics may occur in self-managingteams in which trust is high that will not occur ineither self-managing teams in which trust is loweror in manager-led teams or work groups. Theseunique dynamics will occur as individual teammembers struggle with the decision about whetheror not to suggest (or support) greater monitoring intheir team. Factors such as the desire to be per-ceived as a “team player” and to conform, the fearof sanction or punishment, and concern for thefeelings of fellow team members will all influencethe aggregate decision of a self-managing team aseach member struggles with these factors. A man-ager deciding on how much to monitor a compara-ble work group, on the other hand, would not besusceptible to these pressures.

FIGURE 1Overall Model

386 JuneAcademy of Management Journal

In both self-managing teams and manager-ledwork groups, low trust will generally be associatedwith higher monitoring, and high trust will be as-sociated with lower monitoring (that is, they willhave a negative relationship). However, the abovemechanisms suggest an additional reluctanceamong members of a self-managing team to imple-ment and enforce monitoring. I would expect thisreluctance to grow as trust increases, and the cor-responding potential costs and risks of suggestingor endorsing monitoring will also become greaterand more salient to team members. Thus, while arelatively linear negative relationship betweentrust and monitoring is expected in a manager-ledteam or work group, in a self-managing team, therelationship is likely to be nonlinear. The level ofmonitoring is expected to drop off more rapidlythan in a manager-led work group as trust increasesand the pressures on team members to not monitorone another grow.

In other words, the higher the level of trust, themore reluctance there will be on the part of mem-bers of a self-managing team to express their desirefor monitoring in the team, and instead there willbe a tendency to suppress the public expression ofthat desire. In combination with conformity andgroupthink pressures, these tendencies will lead toless monitoring at higher levels of trust than whatone would normally expect in a manager-led teamor work group.

At relatively low levels of trust, however,there is little reason to expect that members of aself-managing team will be similarly reluctant toexpress their desire for monitoring. Under condi-tions of low trust, a team likely has little cohesive-ness and power to collectively sanction or punishindividuals for suggesting greater monitoring, andteam members are much less likely to perceive atrust violation if trust is already low. The mecha-nisms that can suppress team members’ expressionof their desire for monitoring when trust is highwill not be as effective when trust is lower. Thus,there is likely to be little difference in the amountsof monitoring implemented by a self-managingteam in which trust is low and a manager-led teamor work group. In fact, Barker (1993) found thatself-managing teams could become quite draconianin their monitoring and controlling behavior whentrust was perceived to be low.

While the mechanism is rooted in individual per-ception and behavior, the outcome is clearly ateam-level phenomenon (as with other team phe-nomena, like groupthink and conformity). The sup-pression of monitoring that is likely only whentrust is higher suggests that the linear relationshipone might expect between trust and monitoring in

a manager-led work group is not going to occur ina self-managing team. Rather, the relationshipshould be nonlinear, albeit still negative. The non-linear relationship that best captures expectationsbased on the arguments above is a predominantlynegative, concave, downward curve.

Hypothesis 1. The relationship between thelevel of intrateam trust and the amount ofmonitoring reported in a team will be negativeand nonlinear.

In the management-employee relationship, somemonitoring is always necessary. According to Pow-ell (1996), even when employees can be trustedenough for management to grant them autonomy,there is still a need to monitor those employees.Sabel (1993) pointed out that in addition to reduc-ing the likelihood and possibility of duplicity,monitoring also routinizes contact between parties:“Cooperation is buttressed by sustained contact,regular dialogue, and constant monitoring” (1993:63). According to agency theory, monitoring pro-tects against individuals’ opportunism and self-interest (Eisenhardt, 1989), but stewardship theorysuggests that the assumption of individual oppor-tunism and self-interest is not always appropriate(Davis, Schoorman, & Donaldson, 1997). Al-though some evidence indicates that surveillancenegatively affects individual motivation (Enzle &Anderson, 1993), most research has supportedthe performance benefits of monitoring (Larson &Callahan, 1990).

Monitoring in teams is expected to be similarlyrelated to performance, not only for the above in-dividual-level reasons, which are consistent withreducing the incidence of free riding and socialloafing in teams (Baron, Kerr, & Miller, 1992), butalso because monitoring is expected to increasecoordination and reduce process loss (Saavedra,Earley, & Van Dyne, 1993). The more team mem-bers are aware of each others’ activities, the betterthey can coordinate their work. Ironically, this for-mulation suggests that the “ideal” relationship be-tween trust and monitoring is quite different fromwhat is likely to occur in a self-managing team—namely, that a certain level of monitoring is neces-sary no matter how high the trust within a teambecomes.

This relationship becomes crucial to team perfor-mance when combined with the level of individualautonomy in a team. The more individual auton-omy there is in a team, the more team members willbe working independently of one another, and themore monitoring and communication will be nec-essary to avoid potential coordination and processlosses (Orton & Weick, 1990). Individual autonomy

2004 387Langfred

is generally expected to have a positive effect onperformance, on the basis of Hackman and Old-ham’s (1976) job characteristics model, and thelogic that decision-making power should be in thehands of the individuals with most informationabout a task (Locke & Schweiger, 1979). However, Isuggest that such a positive relationship is contin-gent on the level of monitoring within a team andthat the autonomy-performance relationship canbecome negative if monitoring is insufficient. Inother words, high levels of individual autonomy ina team should be accompanied by relatively highlevels of monitoring, and insufficient monitoringcould lead to lower performance. This statementimplies the following interaction between monitor-ing and autonomy:

Hypothesis 2. Individual autonomy and moni-toring will interact in such a way that teamswith high individual autonomy and low mon-itoring will have lower performance thanteams with high individual autonomy and highmonitoring.

The logic outlined above in Hypotheses 1 and 2suggests that monitoring mediates the effect of truston performance. In other words, if trust affectsmonitoring, and monitoring affects performance,trust and performance must have an indirect rela-tionship, carried (in part or in full) by monitoring.The fact that the relationship between monitoringand performance is also expected to depend on thelevel of autonomy (the interaction in Hypothesis 2)complicates the mediation. The overall model, asshown in Figure 1, describes “moderated media-tion” (Baron & Kenny, 1986; James & Brett, 1984) inwhich the relationship between the mediator andthe outcome variable is moderated by another vari-able. In this particular case, trust affects monitoring(the mediator), which interacts with autonomy (themoderator) to affect performance. To support theoverall model, it is therefore necessary to establishthis moderated mediation and show that the indi-rect effect of trust on performance is carried throughmonitoring and its interaction with autonomy.

METHODS

Setting

The research participants were two complete co-horts of MBA graduate students organized in self-managing teams at the business school of a privatemidwestern university. Students were assigned toteams and worked in them for four months, toperform a wide variety of tasks (including financialanalyses, marketing projects, statistical problem

sets, business case write-ups, presentations, andlonger papers and projects) pertaining to eightclasses. During a week of orientation, team mem-bers performed a number of team-building activi-ties, followed immediately by the start of classes.The teams were self-managing and had completediscretion in deciding how to carry out assign-ments. Teams were not formally evaluated on theirprocesses or methods for carrying out assignments,but only on the quality of their output. Team place-ments were based on criteria designed to maximizeheterogeneity on gender, nationality, educationalbackground, and work experience. Every team hadfour members, both genders, at least two national-ities, and at least three different undergraduate ma-jors. This even distribution provided a limitedmethodological control for some demographic vari-ables in that it reduced the need to include thosevariables as controls in the statistical analyses andfurther limit degrees of freedom.

Upon completion of classes, all teams were re-quired to participate in a series of four case com-petitions. They were given 24 hours after receivingeach case to create analyses and recommendationsto present to a panel of faculty and industry ex-perts, who evaluated their performance. The casecompetition was a central part of the school’s MBAcurriculum, and students were told throughout thepreceding semesters that it was the culmination ofthe required coursework. Winners of the competi-tion enjoyed prestige and recognition, in additionto various prizes. The data were collected duringthis week of case competition for each class, andthe questionnaire specifically referred to teams’activities during the competition. Teams weredropped from the analysis if fewer than half theteam members responded to the surveys.

Participants

There were 300 participants in 76 teams. Datafrom 35 teams were collected from one incomingclass of 135 students, and data from 41 additionalteams were collected from the following class of165 students. A response rate of 83 percent yielded248 respondents, representing 71 teams. Of the re-spondents, 21.6 percent were women, and 78.4 per-cent were men. U.S. citizens comprised 66.5 per-cent of the respondents, with the largest non-U.S.contingents made up of Chinese, Indian, Japanese,and Korean nationals (in order of representation).The average age was 29.4 years; the youngest indi-vidual was 22, and the oldest was 47. Undergrad-uate degrees spanned a variety of majors, with themost common being business, social science, engi-neering, and economics. Participation was com-


pletely voluntary and was not linked to coursecredit or financial reward. There were no signifi-cant differences between the two cohorts of MBAstudents on any variables, and the case competi-tion, timing, data collection, and measures wereidentical for both cohorts.

Measures

Measures in this study consisted of performancemeasures rated by a panel of experts, survey ques-tionnaires filled out by team members, and archivalindividual data.

Individual ability. This was measured by stu-dents’ GMAT scores, which were obtained fromschool records, not self-reported.

Team performance. This was measured by per-formance during the case competition and was theaverage of the numerical ratings made by a mini-mum of six raters who witnessed the presentationof teams’ case analyses and had the opportunity toquestion the teams. The raters were faculty, indus-try experts, former students, and communicationspecialists. Each rater assigned points on six differ-ent dimensions of a team’s performance, includingthe analysis, presentation, and the performance ofthe team in a question-and-answer session.

The survey questionnaire assessed a number ofconstructs using multi-item scales. Appendix Agives the scale items and factor analysis results.

Intrateam trust. This four-item scale, based onSimons and Peterson’s (2000) scale, had a Cron-bach alpha of .83 and a mean intraclass correlationcoefficient (ICC) of .80.

Individual autonomy. This three-item measure,based on Breaugh’s (1989) scale, had a Cronbachalpha of .84 and a mean ICC of .81. This measure isthe mean level of individual autonomy reported byindividual team members.

Monitoring. This four-item measure, based onCummings and Bromiley’s (1996) scale, had a Cron-bach alpha of .81 and a mean ICC of .85.

Analysis

To test the interaction effect hypothesized, I usedthe multiplicative product of the variables in hier-archical multiple regression analyses, as Baron andKenny (1986) recommended. Change in the amountof variance explained (�R2) is the most appropriatetest of the significance of an interaction term (Co-hen & Cohen, 1983). Interactions were plotted byderiving separate equations for the high and low(one standard deviation above and below the mean)conditions, as recommended by Aiken and West(1991). Because the regression analyses involved

interactions, the “main effect” terms and productterms could be highly correlated, raising the issueof multicollinearity, which can make regression co-efficients unstable and difficult to interpret (Cohen& Cohen, 1983). Variables in the study were thuscentered to reduce multicollinearity (Aiken & West,1991).

The moderated mediation underlying the modelwas tested via a series of hierarchical regressionsbased on the four steps recommended by Baron andKenny (1986). The goal of step 1 was to establishthe relationship between trust and performance inthe absence of monitoring (the mediator). In thesame manner that monitoring and autonomyshould interact to influence performance, trust andautonomy should interact when monitoring is notincluded in the model. The second step was dem-onstrating the relationship between trust and mon-itoring. The third step was establishing the rela-tionship between monitoring and performance (inthis case, also interacting with autonomy), and thefourth step was demonstrating that the effect of theinitial variable (the interaction of trust and auton-omy) was reduced or insignificant when the medi-ator (the interaction of monitoring and autonomy)was added to the model. Appendix B discusses thisanalysis in greater detail.

RESULTS

The values of the intercorrelation coefficientsconfirmed the appropriateness of aggregation forintrateam trust, monitoring, and individual auton-omy. As a check for multicollinearity, variance in-flation factor (VIF) scores were calculated for thevariables in each regression model. All VIF scoreswere below 4, and most were below 2, suggestingthat multicollinearity was not a serious problem inthe analysis.

Table 1 provides means, standard deviations,scale reliabilities, and correlations.

Hypothesis 1 predicts a negative, nonlinear rela-tionship between the levels of trust and monitoringin teams. Since testing this relationship was part ofthe mediation analysis (Baron & Kenny, 1986), theresults are shown in Table 2 under “Mediation Step2.” The relationship shown there is significant andnegative (t68 � �8.35, p � .01), establishing a neg-ative relationship between trust and monitoring.(This result also satisfied step 2 of the four-stepmediation analysis.)

Demonstrating the nonlinear (concave down-ward) relationship required two things. First, anadditional regression equation (or hierarchicalstep) had to be run with the trust variable squaredadded. The beta weight had to be significant, and

2004 389Langfred

the change in variance explained from one equa-tion (step) to the next also had to be significant.This is also shown in “Mediation Step 2” in Table2, with a significant beta weight (t67 � �2.19, p �.05) and a significant change in the explained vari-ance (p � .05) Second, to confirm the existence of adownward concave curve (as opposed to anothercurvilinear relationship), both the coefficients ofthe main term and of the squared term had to beshown to be negative (Aiken & West, 1991). Theresults also confirm this condition. The data plot-ted in Figure 2 confirm that (1) there is a nonlinearrelationship, and (2) the relationship is a concavedownward curve (the shaded line illustrates thenonlinear trend of the data). Overall, there is sup-port for Hypothesis 1.

Hypothesis 2 predicts that autonomy and moni-toring interact in such a way that teams with highlevels of individual autonomy and low levels ofmonitoring will perform worse than teams withhigh levels of autonomy and high levels of moni-toring. The testing of Hypothesis 2 was step 3 of themediation test (Baron & Kenny, 1986). The resultsare illustrated in “Mediation Step 3 and 4” in Table2, in which the monitoring by autonomy term issignificant (t63 � 2.65, p � .05), with trust, moni-toring, autonomy, and the trust by autonomy inter-action controlled. If a change in explained variance(�R2) is calculated for the addition of the monitor-ing by autonomy interaction term, it is also signif-icant (�F64, 1 � 7.01, p � .05), further supportingHypothesis 2 (and establishing the third step of themediation analysis).

The first step of the mediation is established in“Mediation Step 1” in Table 2, as the interaction oftrust and autonomy has a significant effect on per-formance (t66 � �2.87, p � .01; �F66, 1 � 8.21, p �.05). The test of Hypothesis 2 (“Mediation Step 3and 4”) also establishes the fourth step of the me-diation analysis, since the trust by autonomy inter-action from the first step is no longer significant

when it is included together with the monitoring byautonomy variable. In other words, when the effectof the trust by autonomy interaction on perfor-mance is explored without the mediating effect ofmonitoring, it is significant, but the effect is elimi-nated when the mediator (the interaction of monitor-ing by autonomy) is added. Thus, these findings,combined with those described above for Hypotheses1 and 2, satisfy all four steps of the mediator analysisrecommended by Baron and Kenny (1986), and itappears that the expected mediation does occur.

Finally, to explore Hypothesis 2 fully, I plottedthe interaction between monitoring and autonomy,as is shown in Figure 3a. The figure shows thatwhen monitoring is lower, the negative relation-ship between autonomy and performance is stron-ger than it is when the level of monitoring is higher.The plot also indicates that teams with low levelsof monitoring and high levels of autonomy per-formed worst.

As shown in “Mediation Step 1,” trust and au-tonomy also have a significant interaction when themediator is not included in the analysis. The inter-action, plotted in Figure 3b, shows the expectedmoderating effect of trust on the relationship be-tween autonomy and performance—mirroring therelationship seen with monitoring and autonomy.

Overall, the results demonstrate strong supportfor both hypotheses and for the overall expectationthat monitoring would mediate the relationship be-tween trust and performance in self-managing teams.

DISCUSSION

Conclusions

The findings have shown that, under some con-ditions, too much trust in a self-managing team canbe harmful. This effect is illustrated when the trust-performance relationship is explored in combina-tion with the level of individual autonomy in a

TABLE 1Means, Standard Deviations, Scale Reliabilities, and Correlationsa

Variable Mean s.d. 1 2 3 4 5 6

1. Age 29.55 1.90 n.a.2. GMATb 0.0 1.0 �.09 n.a.3. Performance 140.03 15.48 .01 .05 n.a.4. Autonomy 6.40 1.02 �.07 .10 �.52** .845. Trust 7.05 1.11 �.08 �.01 �.10 .07 .836. Monitoring 5.04 1.36 .09 .09 .14 �.09 �.71** .81

a n � 71.b Graduate Management Achievement Test; standardized score.** p � .01


team. As it turns out, high levels of individualautonomy can become a liability in self-managingteams when the level of trust is high and littlemonitoring takes place. The mediator analysisdemonstrated that the indirect effect of trust ap-pears to be accounted for by the level of monitoringin a team. This pattern suggests that the more teammembers trust one another, the less they chose to

monitor one another, and when this condition iscombined with high levels of individual autonomy,performance suffers.

The finding that trust can lead to a performanceloss is counterintuitive in that trust has tradition-ally been regarded as a benefit to teams and organ-izations. It is important to note that the presentfinding is an interaction effect: too much trust washarmful only in self-managing teams characterizedby high levels of individual autonomy. Generally,the conventional benefits of trust are still expectedto hold, and I do not intend to suggest otherwise.The findings are, however, particularly importantto the trust literature, because negative effects ofhigh trust on performance have not previously beenempirically explored, even though several re-searchers have suggested that high trust could havea downside (Kramer, 1999; McEvily, Perrone, &Zaheer, 2003).

The practical implication of the findings is notthat trust should be avoided in self-managingteams. Rather, the implication is that if a team hashigh levels of individual autonomy, some monitor-ing of individual team members needs to be inplace if process loss and coordination errors are tobe avoided. In self-managing teams, this appears tobe particularly important, as high levels of in-trateam trust are especially likely to make teammembers reluctant to monitor one another. Thepractical implication is essentially that a lack ofmonitoring can be naı̈ve, regardless of levels oftrust, and that a little skepticism never hurt any-one—or any team.

The contribution of these findings extends be-yond the counterintuitive conclusion that trust canbe harmful in teams, given certain circumstances.These findings also emphasize the importance ofexploring moderators in trust research, particularlyin team settings (Dirks & Ferrin, 2001), and theydefine some important boundary conditions for thebenefits of trust. In addition, the findings help tolink the literature of trust with that of autonomyand self-management in teams, and they expandthe scope of current trust research by exploring theeffects of the interaction of trust and team charac-teristics (like autonomy) on team performance.

Limitations and Future Research

A restriction of range may exist in the data, inthat levels of trust and autonomy were generallyhigh in most of the studied teams, reducing desiredvariability. The study essentially explored the dif-ference between moderate and very high levels oftrust, as opposed to the difference between low andhigh levels of trust. However, such a restriction of

TABLE 2Moderated Mediation: Results of Hierarchical

Multiple Regression Analysisa

Steps and Variables � F R2 �R2 VIF

Mediation step 1: Performance

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01

Step 2Intrateam trust �.06 6.45** .28 .28** 1.02Autonomy �.52** 1.02

Step 3Trust � autonomy �.29** 7.36** .36 .08** 1.06

Mediation step 2:Monitoring, Hypothesis1

Step 1Age .10 0.62 .02 1.01GMAT .10 1.01

Step 2Intrateam trust �.71** 24.06** .52 .50** 1.01

Step 3Intrateam trust squared �.22* 20.26** .55 .03* 1.54

Mediation step 3 and 4:Performance,Hypothesis 2

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01

Step 2Intrateam trust �.01 4.27** .28 .28** 3.06Autonomy �.51** 1.04Monitoring .10 2.22Intrateam trust squared .05 1.68

Step 3Intrateam trust �

autonomy.05 6.14** .44 .16** 3.32

Monitoring � autonomy .49* 3.79

a Each mediation step contains a hierarchical regression anal-ysis having three steps.

* p � .05** p � .01

2004 391Langfred

range would normally create a problem for analysisand causal inference, not for generalizability. Thefact that effects were found, including interactions,suggests relative robustness. In terms of generaliz-ability, the composition of the teams may be alimiting factor. Although the teams were quite di-verse (in terms of national origin, gender, educa-

tion, and so forth) they could also be perceived asfairly homogeneous, since all team members wererelatively close in age and were MBA students atthe same university.

A possible concern surrounding the relationshipbetween trust and monitoring is the direction ofcausality. The use of cross-sectional data precludes

FIGURE 2Relationship Between Trust and Monitoring in Self-Managing Teams

FIGURE 3Interaction Effects


drawing inferences about causal direction. How-ever, the fact that team members had worked to-gether on a variety of tasks for many months sug-gests that fairly stable perceptions of trust in otherteam members were established. Since the task in-volved in the study was novel and accompanied bysevere time pressure (unlike any other task theteams had performed before), it is likely that theteams developed new techniques, roles, and coor-dination (including new levels of monitoring) forthis task.

One exciting avenue for future research is therole of task interdependence, and the question ofwhether or not it might provide additional bound-ary conditions for the relationships revealed in thisstudy. The relationship between individual auton-omy, monitoring, and performance is in part de-pendent on coordination, communication, and mu-tual adjustment within a team; these are alsoimportant variables in the considerable literatureon task interdependence and its relationship withperformance (Saavedra et al., 1993).

Finally, while my focus has been on one partic-ular boundary condition for the benefits of trust inteams, the findings should not be viewed as sup-porting one counterintuitive aberration, but ratheras a general direction for the future. As the organi-zational research on trust continues to grow, it willbe important to balance the study of the benefits oftrust with the acknowledgment that those benefitshave limits. I have shown that in self-managingteams characterized by high levels of individualautonomy, too much trust can harm team perfor-mance. It is likely that there are other conditionsunder which trust may be harmful, or at least maynot have the positive effects normally associatedwith it. Such examples might be found in the con-text of cultural differences, negotiation and conflictresolution, team design, or in a variety of otherresearch areas.

REFERENCES

Aiken, L., & West, S. 1991. Multiple regression: Testingand interpreting interactions. Newbury Park, CA:Sage.

Barker, J. R. 1993. Tightening the iron cage: Concertivecontrol in self-managing teams. Administrative Sci-ence Quarterly, 38: 408–437.

Baron, R. M., & Kenny, D. A. 1986. The moderator-medi-ator variable distinction in social psychological re-search: Conceptual, strategic and statistical con-siderations. Journal of Personality and SocialPsychology, 51: 1173–1182.

Baron, R. S., Kerr, N. L., & Miller, N. 1992. Group process,

group decision, group action. Pacific Grove, CA:Brooks/Cole.

Baron, R. S., Vandello, J. A., & Brunsman, B. 1996. Theforgotten variable in conformity research: Impact oftask importance on social influence. Journal of Per-sonality and Social Psychology, 71: 915–927.

Breaugh, J. A. 1989. The work autonomy scales: Addi-tional validity evidence. Human Relations, 42:1033–1056.

Brockner, J., Chen, Y. R., Mannix, E. A., Leung, K., &Skarlicki, D. P. 2000. The moderating influence ofcultural differences in self-construal on the interac-tive relationship between outcome favorability andprocedural fairness. Administrative Science Quar-terly, 45: 138– 159.

Bromiley, P., & Cummings, L. L. 1995. Transactions costsin organizations with trust. In R. Bies, B. Sheppard,& R. Lewicki (Eds.), Research on negotiation inorganizations, vol. 5: 219– 247. Greenwich, CT: JAI.

Cohen, J., & Cohen, P. 1983. Applied multiple regres-sion/correlation analysis for the behavioral sci-ences (2nd ed). Hillsdale, NJ: Erlbaum.

Cohen, S. G., Ledford, G. E., & Spreitzer, G. M. 1996. Apredictive model of self-managing work team effec-tiveness. Human Relations, 49: 643–676.

Collins, L. M., Graham, J. W., & Flaherty, B. P. 1998. Analternative framework for defining mediation. Mul-tivariate Behavioral Research, 33: 295–312.

Creed, W. E., & Miles, R. E. 1996. Trust in organizations:A conceptual framework linking organizationalforms, managerial philosophies, and the opportunitycosts of control. In R. M. Kramer & T. R. Tyler (Eds.),Trust in organizations: Frontiers of theory and re-search: 16–39. Thousand Oaks, CA: Sage.

Cummings, L. L., & Bromiley, P. 1996. The Organiza-tional Trust Inventory (OTI): Development and vali-dation. In R. M. Kramer & T. R. Tyler (Eds.), Trust inorganizations: Frontiers of theory and research:302–330. Thousand Oaks, CA: Sage.

Davis, J. H., Schoorman, F. D., & Donaldson, L. 1997.Towards a stewardship theory of management.Academy of Management Review, 22: 20–47

Deutsch, M. 1958. Trust and suspicion. Journal of Con-flict Resolution, 2: 265–279

Dirks, K. T. 1999. The effects of interpersonal trust onwork group performance. Journal of Applied Psy-chology, 84: 445–455.

Dirks, K. T., & Ferrin, D. L. 2001. The role of trust inorganizational settings. Organization Science, 12:450–467.

Eisenhardt, K. M. 1989. Agency theory: An assessmentand review. Academy of Management Review, 14:57–74.

Enzle, M. E., & Anderson, S. C. 1993. Surveillant inten-

2004 393Langfred

tions and intrinsic motivation. Journal of Personal-ity and Social Psychology, 64: 257–266.

Feldman, D. C. 1984. The development and enforcementof group norms. Academy of Management Review,9: 47–53.

Friedlander, F. 1970. The primacy of trust as a facilitatorof further group accomplishment. Journal of Ap-plied Behavioral Science, 6: 387–400.

Hackman, J. R. 1983. Designing work for individuals andfor groups. In J. R. Hackman (Ed.), Perspectives onbehavior in organizations: 242–256. New York:McGraw-Hill.

Hackman, J. R., & Oldham, G. R. 1976. Motivationthrough the design of work: Test of a theory. Or-ganizational Behavior and Human Performance,16(2): 250–279.

Hull, J. G., Tedlie, J. C., & Lehn, D. A. 1992. Moderatorvariables in personality research: The problem ofcontrolling for plausible alternatives. Personalityand Social Psychology Bulletin, 18: 115–117.

James, L. R., & Brett, J. M. 1984. Mediators, moderators,and tests for mediation. Journal of Applied Psychol-ogy, 69: 307–321.

Janis, I. L. 1982. Groupthink (2nd ed.). Boston: HoughtonMifflin.

Kegan, D. L., & Rubinstein, A. H. 1973. Trust, effective-ness, and organizational development: A field studyin R&D. Journal of Applied Behavioral Science,vol. 9: 498–513.

Kenny, D. A., Kashy, D. A., & Bolger, N. 1998. Dataanalysis in social psychology. In D. Gilbert, S. Fiske,& G. Lindzey (Eds.), The handbook of social psy-chology, vol. 1 (4th ed.): 233–265. Boston: McGraw-Hill.

Klein, K. J., Dansereau, F., & Hall, R. J. 1994. Levels issuesin theory development, data collection, and analysis.Academy of Management Review, 19: 195–229.

Korsgaard, M. A., Brodt, S. E., & Whitener, E. M. 2002.Trust in the face of conflict: The role of managerialtrustworthy behavior and organizational context.Journal of Applied Psychology, 87: 312–319.

Kramer, R. M., & Tyler, T. R. 1996. Trust in organiza-tions: Frontiers of theory and research. ThousandOaks, CA: Sage.

Kramer, R. M. 1999. Trust and distrust in organizations:Emerging perspectives, enduring questions. In J. T.Spence, J. M. Darley, & D. J. Foss (Eds.), Annualreview of psychology, vol. 50: 569–598. Palo Alto,CA: Annual Reviews.

Langfred, C. W. 2000. The paradox of self-management:Individual and group autonomy in work groups.Journal of Organizational Behavior, 21: 563–585.

Larson, J. R., & Callahan, C. 1990. Performance monitor-ing: How it affects work productivity. Journal ofApplied Psychology, 75: 530–538.

Lewicki, R. J., & Bunker, B. B. 1996. Developing andmaintaining trust in work relationships. In R. M.Kramer & T. R. Tyler (Eds.), Trust in organizations:Frontiers of theory and research: 114–139. Thou-sand Oaks, CA: Sage.

Locke, E. A., & Schweiger, D. M. 1979. Participation indecision-making: One more look. In B. Staw (Ed.),Research in organizational behavior, vol. 1: 265–340. Greenwich, CT: JAI Press.

MacKinnon, D. P. 2000. Contrasts in multiple mediatormodels. In J. S. Rose, L. Chassin, C. C. Presson, & S. J.Sherman (Eds.), Multivariate applications in sub-stance use research: New methods for new ques-tions: 141–160. Mahwah, NJ: Erlbaum.

Mayer, R. C., Davis, J. H., & Schoorman, F. D. 1995. Anintegrative model of organizational trust. Academyof Management Review, 20: 709–734.

McEvily, B., Perrone, V., & Zaheer, A. 2003. Trust as anorganizing principle. Organization Science, 14: 91–103.

Orton, J. D., & Weick, K. E. 1990. Loosely coupled sys-tems: A recoceptualization. Academy of Manage-ment Review, 15: 203–223.

Powell, W. W. 1996 Trust-based forms of governance. InR. M Kramer & T. R. Tyler (Eds.), Trust in organiza-tions: Frontiers of theory and research: 51-67.Thousand Oaks, CA: Sage.

Rousseau, D. M. 1985. Issues of level in organizationalresearch: Multi-level and cross-level perspectives. InL. L. Cummings & B. M. Staw (Eds.), Research inorganizational behavior, vol. 7: 1–37. Greenwich,CT: JAI Press.

Saavedra, R., Earley, P. C., & Van Dyne, L. 1993. Complexinterdependence in task-performing groups. Journalof Applied Psychology, 78: 61–72.

Sabel, C. F. 1993. Constitutional ordering in historicalcontext. In F. W. Scharpf (Ed.), Games in hierar-chies and networks: 65–123. Boulder, CO: West-view.

Sheeran, P., & Abraham, C. 2003. Mediator of modera-tors: Temporal stability of intention and the inten-tion-behavior relation. Personality and Social Psy-chology Bulletin, 29: 205–215.

Shrout, P. E., & Bolger, N. 2002. Mediation in experimen-tal and nonexperimental studies: New proceduresand recommendations. Psychological Methods, 7:422–445.

Shrout, P. E., & Fleiss, J. L. 1979. Intraclass correlations:Uses in assessing rater reliability. PsychologicalBulletin, 86: 420–428.

Simons, T. L., & Peterson, R. S. 2000. Task conflict andrelationship conflict in top management teams: Thepivotal role of intragroup trust. Journal of AppliedPsychology, 85: 102–111.

Wellins, R. S., Wilson, R., Katz, A. J., Laughlin, P., Day,


C. R., & Price, D. 1990. Self-directed teams: A studyof current practice. Pittsburgh: DDI.

APPENDIX A

Questionnaire Items and Factor Analysis Results

All questions were asked with respect to the specificcontext of the case competition, with response optionsranging from 1, “strongly disagree,” to 9, “strongly agree”.

Individual Autonomy

1. In the team, I decide how to do my own work.2. On team projects, I control the scheduling of my work.3. Once the team decides what to do, I decide how to do

my part.

Trust

1. We trust each other a lot in my team.2. I know I can count on the other team members.3. The other team members know they can count on me.4. I trust all of the other team members.

Monitoring

1. We check to make sure that other team members con-tinue to work on team projects.

2. We monitor each others progress on team projects.3. We check whether everybody is meeting their obliga-

tion to the team.4. We watch to make sure everyone in the team meets

their deadlines.

APPENDIX B

Analysis of Moderated Mediation: Explanation andAlternatives

Conceptualizing Moderated Mediation

The purpose of this appendix is to explain the analysesinvolved in the moderated mediation analysis of thetheoretical model (Figure 1) in the paper. From an ana-lytic viewpoint (that is, constructing regression equa-tions), however, this model is actually no different fromthe model shown in Figure B1a, since “moderation” rep-resents the interaction between two constructs. Whilewhich one moderates which other one is meaningfultheoretically, it is irrelevant as far as the analysis (spe-cifically, the multiplicative term in a regression equa-tion) is concerned.

Both models show the same moderated mediation inwhich the mediator (monitoring) interacts with anothervariable (autonomy) to cause the outcome variable (per-formance, which will be referred to as “y” in this appen-dix). It is also worth noting that there are two differenttypes of moderated mediation (James & Brett, 1984;Kenny et al., 1998), each requiring different analyses. Inone of these (referred to as type 1 here), the moderationoccurs between the initial variable and the mediator, andin the other (here, type 2), the moderation occurs be-tween the mediator and the outcome variable. The anal-ysis of type 2 is more complex, in that only one interac-tion term is necessary for a type 1 analysis (as isillustrated in Korsgaard, Brodt, and Whitener [2002]),whereas two interactions may be required for a type 2analysis. My model is clearly a type 2.

To be specific (and just to complicate matters further),this particular model actually represents “curvilinearmoderated mediation,” since the relationship betweenthe initial variable (trust) and the mediator (monitoring)was expected to be curvilinear. Fortunately, this addi-tional aspect has no effect on the analyses beyond theneed to include trust squared in some of the equations. (Ifthe relationship were linear, the squared term would notappear in any of the equations.)

Before discussing the relatively complex case of mod-erated mediation analysis, it is worth reviewing Baronand Kenny’s (1986) procedure for nonmoderated, “sim-ple” mediation analysis, (that is, the case in which mon-itoring would mediate the effect of trust on y, but nomoderator would be involved). For the variables of inter-est here, Figure B2a illustrates a simple mediation modelthat is carried out in four steps.

Step 1—Establish the relationship of trust with y: y � f(trust).

Step 2—Establish the relationship of trust with monitor-ing: monitoring � f (trust, trust squared)

Step 3—Establish the relationship of monitoring with y,controlling for trust: y � f (monitoring, trust,trust squared)

Step 4—Establish that the effect of trust on y is elimi-nated (showing full mediation) or reduced (par-tial mediation) when monitoring is in the same

Results of TABLE A1Factor Analysisa

Scales and Items

Loadings

Factor 1 Factor 2 Factor 3

Autonomy 1 .01 �.06 .88Autonomy 2 .02 �.07 .89Autonomy 3 .01 .02 .85

Trust 1 .88 �.06 .01Trust 2 .90 �.07 .01Trust 3 .68 �.19 �.03Trust 4 .79 .13 .13

Monitor 1 �.03 .77 .08Monitor 2 .03 .85 .02Monitor 3 �.07 .84 �.07Monitor 4 �.09 .75 �.14

a Both a scree plot and an eigenvalue cutoff of 1.0 yieldedthree factors. Significant loadings are shown in bold.

2004 395Langfred

equation. This is accomplished with the step 3equation.

Moderated mediation, which was hypothesized in thisstudy, is represented in Figure B2b, illustrating the stepsof the mediation analysis. Here, the mediator (monitor-ing) interacts with another variable (autonomy) to causethe outcome variable (y):

Alternative Procedures for Testing ModeratedMediation

Unfortunately, the analytical procedure for this mod-erated mediation model is not obvious. I have outlinedthe three most plausible alternatives below, labelingthem options 1, 2, and 3. The main difference betweenoption 1 and option 2 is whether or not the trust byautonomy interaction appears only in step 3, or in bothsteps 1 and 3. Option 3 extends the classic mediationanalysis by adding a separate equation to test step 4.

Option 1. This method does not include the interactionterm in step 1.

Step 1—Establish the relationship of trust with y: y � f(trust).


Step 3—Establish the relationship of the interaction ofmonitoring and autonomy with y, controlling fortrust: y � f (trust, trust squared, monitoring,autonomy, monitoring � autonomy)

Step 4—Establish that the effect of trust on y is elimi-nated (showing full mediation) or reduced (par-tial mediation) when monitoring and autonomyare in the same equation. This is accomplishedwith the step 3 equation.

The results are shown in Table B1.

While this may seem to be the most intuitive or obvi-ous extension of a classic Baron and Kenny (1986) me-

FIGURE B2Models of Moderated Mediation Analysis

FIGURE B1a

Models of Moderated Mediation Analysis

aPerformance is represented by “y.”


diator analysis, step 1 is problematic. Specifically, ifmediation takes place, and especially if the interactionbetween monitoring and autonomy is disordinal (whichit turned out to be in this case), then there is no reason toexpect a direct (nonmoderated) link between trust and yin step 1. If the effect of trust on y is carried through themediator, then it is also carried through that mediator’sinteraction—and one wouldn’t expect a detectable directeffect of trust on y. Thus, if partial or full mediation werepresent, it would be unreasonable to expect step 1 of thismediation analysis to be satisfied.

Recently it has been suggested that step 1 of the classicmediation analysis is not actually necessary to establishmediation (Collins, Graham, & Flaherty, 1998; MacKin-non, 2000; Shrout & Bolger, 2002). The logic is that step1 is not required, since a path from the initial variable tothe outcome is implied if steps 2 and 3 are met. Theemphasis is on steps 2 and 3 over step 1, since there area variety of ways in which mediation could be occurring,

but the direct relationship in step 1 still would not besignificant—including causal distance, suppressor vari-ables, and (as is the case in this particular model), con-tingencies operating on the mediator itself. This viewsuggests that option 1 may in fact be a viable method fortype 2 moderated mediation analysis, with the caveatthat step 1 does not have to be significant. A slightvariation on option 1 would be to include the autonomyterm in step 1, consistent with the implied logic of Jamesand Brett (1984), as well as the empirical analysis byLangfred (2000).

Option 2. This method acknowledges the problemwith option 1 noted above and includes the interaction oftrust and autonomy in step 1. This method relies on theassumption that the step 1 relationship has to be moder-ated if the mediator’s relationship with the outcome vari-able is itself moderated, as illustrated in Figure B2c.

The regression analysis takes the following form:

Step 1—Establish the relationship of the interaction oftrust and autonomy with y: y � f (trust, auton-omy, trust � autonomy). This the relationshipthat trust has with y in the absence of monitor-ing.


Step 3—Establish the relationship of the interaction ofmonitoring and autonomy with y, controlling forthe interaction of trust and autonomy: y � (trust,trust squared, autonomy, monitoring, trust �autonomy, monitoring � autonomy)

Step 4—Establish that the effect of the interaction of trustand autonomy on y is eliminated (full media-tion) or reduced (partial mediation) when mon-itoring is in the same equation. This is accom-plished with the step 3 equation.

The steps just outlined compose the method used inthe study, and the results are shown in Table 2. Hull,Tedlie, and Lehn’s (1992) discussion of controlling formoderators implies support for this method. Empirically,Brockner, Chen, Manniz, Leung, and Skarlicki (2000)included both interaction terms in their analysis, as didSheeran and Abraham (2003).

This method raises a different question, however. Log-ically, this set of regression equations seems to imply thatthe interaction between monitoring and autonomy medi-ates the effect of the interaction of trust and autonomy ony, which is not precisely what my theoretical modelindicates. Rather, my model suggests that trust and au-tonomy interact, in the absence of other variables, toaffect y. When monitoring is introduced into the picture,it turns out that it is monitoring that interacts with au-tonomy, and it only appeared to be trust because trust ishighly correlated with monitoring. That is a qualitativelydifferent theory than the suggestion that the interactiveeffect of trust and autonomy on y is mediated (or carried)by the interactive effect of monitoring and autonomy ony. My theory posits the type 2 moderated mediationshown as “Model A” in Figure B1a above. The analysis in

TABLE B1Results of Moderated Mediation Analysis, Option 1


Mediation step 1:Performance

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01


Mediation step 2:Monitoring

Step 1Age .10 0.62 .02 1.01GMAT .10 1.01

Step 2Intrateam trust �.71 24.06** .52 .50** 1.01

Step 3Intrateam trust squared �.22 20.26** .55 .03* 1.54

Mediation step 3 and 4:Performance

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01


Step 3Monitoring � autonomy .44** 7.11** .44 .16** 1.28

2004 397Langfred

option 2, however, seems to test the “Model B” shown inFigure B1b, which includes both type 1 and type 2 mod-erated mediation:

That interpretation is supported by the work of Jamesand Brett (1984), who tested exactly such a model (ModelB) with the regression methodology laid out in option 2.This suggests that the set of regression equations in op-tion 2 might not be entirely appropriate for model A, as ittests a different and slightly more complex model. It isworth noting that Brockner et al. suggested that the mon-itoring by autonomy interaction does not have to besignificant when it is in the same equation as the trust byautonomy interaction to show mediation, just that thereduction or elimination of the trust by autonomy inter-action has to occur. In fact, Brockner et al.’s analysis isconsistent with the final option discussed below, whichis a variation (or extension) of option 2.

Option 3. This method includes the interaction term instep 1 but separates steps 3 and 4 into separate regressionequations, in a departure from Baron and Kenny’s (1986)approach.

Step 1—Establish the relationship of the interaction oftrust and autonomy with y: y � f (trust, auton-omy, trust � autonomy); that is the relationshipthat trust has with y in the absence of monitor-ing.

Step 2—Establish the relationship of trust with monitor-ing: monitoring � f (trust, trust squared).

Step 3—Establish the relationship of the interaction ofmonitoring and autonomy with y, controlling fortrust, but not controlling for the interaction be-tween trust and autonomy: y � (trust, trustsquared, autonomy, monitoring, monitoring �autonomy)

Step 4—Establish that the effect of the interaction of trustand autonomy on y is eliminated (full media-tion) or reduced (partial mediation) when mon-itoring is in the same equation. This is accom-plished with a different equation than the step 3equation, namely the equation with both inter-active terms: y � (trust, trust squared, auton-omy, monitoring, trust � autonomy, monitor-ing � autonomy). However, this equation isused only to test the reduction or elimination ofthe trust � autonomy term, not to test for theeffect of the monitoring � autonomy term on y.

The results of this method are shown in Table B2.

This may be the most appropriate method with whichto analyze type 2 moderated mediation, but it differsfrom the traditional Baron and Kenny (1986) methodol-ogy by adding a fourth equation for step 4. Option 3includes the trust by autonomy interaction in step 1 likeoption 2—on the basis of the same logic that the directrelationship is meaningless—but uses two differentequations for steps 3 and 4. Step 3 tests the effect of themonitoring by autonomy interaction on y, controlling forthe relationship between trust and monitoring, which isnot moderated, and therefore the trust by autonomy termdoes not appear in this equation. If significant, this es-tablishes the mediation. A fourth step is then needed to

TABLE B2Results of Moderated Mediation Analysis Option 3


Mediation step 1:Performance

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01



autonomy�.29** 7.36** .36 .08** 1.06

Mediation step 2:Monitoring

Step 1Age .10 0.62 .02 1.01GMAT .10 1.01

Step 2Intrateam trust �.71 24.06** .52 .50** 1.01

Step 3Intrateam trust squared �.22 20.26** .55 .03* 1.54

Mediation Step 3:Performance

Step 1Age .02 0.08 .00 1.01GMAT .05 1.01


Step 3Monitoring � autonomy .44** 7.11** .44 .16** 1.28

Mediation Step 4:Performance

Step 1Age .02 0.08 .00 1.02GMAT .05 1.02



autonomy.05 6.14** .44 .16** 3.32

Monitoring � autonomy .49* 3.79


explore to what extent the monitoring by autonomy effecton y reduces or eliminates the original effect of the trustby autonomy interaction on y. In step 4, therefore, bothinteractions are included, and the interest is in thechange in significance (if any) between the trust by au-tonomy term in step 1 and in this step, to determinewhether or not the mediation is partial or full. Thus, instep 4, the significance of the monitoring by autonomyterm itself is unimportant.

Conclusion

There is no clear consensus on which of the methodsabove is the correct one. There are proponents of option1, but not all agree that step 1 is unnecessary for testingmediation, and there is some disagreement as to howmany variables to enter in step 1. There is also supportfor option 2, but it seems unlikely that the same meth-odology would be correct to test both the model A andmodel B illustrated above. Thus, it is not clear that option2 is the most correct methodology for type 2 moderatedmediation. Option 3 provides most information by usingfour equations and including all steps from both option 1and 2. Option 3 has support in the empirical literature,just as options 1 and 2 do, but it does depart from thetraditional Baron and Kenny (1986) format by introduc-ing a fourth equation.

As can be seen from the regression results in Tables 2,

B1, and B2, results for each of the three options sup-ported the model. In the article, I chose to report theoption 2 results, as that method is the most conservativein that it includes both interactions in step 3 and requiresthe significance of the monitoring by autonomy interac-tion term as well as the nonsignificance of the trust byautonomy term. For future use, I recommend option 3,since it includes all of the information of option 2, butprovides information beyond that with the additionalstep in the analysis. As such, nothing from option 2 islost by using option 3, and it may provide the mostprudent and comprehensive method for the future anal-ysis of this type of moderated mediation.

Claus W. Langfred ([email protected]) is an asso-ciate professor of organizational behavior at the OlinSchool of Business at Washington University. He re-ceived his Ph.D. from Northwestern University in 1998.His research interests include intrateam trust, cohesive-ness, and conflict, as well as the topics of individualautonomy, team autonomy, and the cross-level implica-tions of studying both simultaneously.

2004 399Langfred

Documents

TOO MUCH OF A GOOD THING? NEGATIVE EFFECTS OF HIGH TRUST AND